LLM outputs are treated as authoritative even when they are ungrounded or incorrect

fluent, confident outputs are treated as authoritative even when ungrounded or incorrect

http://arxiv.org/abs/2608.27102v1